Dialogue Evaluation 2023
RuCoCo
Coreference Resolution
The shared task is over. Thank you all for participating!
Participants may submit a paper for publication in the Conference Proceedings "Computational Linguistics and Intellectual Technologies" with a description of the solution and analysis of the results. Deadline for submission of Dialogue Evaluation articles: 8 April 11:59 pm. Please read the publishing rules for the Dialogue Evaluation track.
Links
Key dates
- 13 January — publication of the train, dev and test datasets;
- 23 March 23:59 (GMT +3) — public phase deadline;
- 26 March 23:59 (GMT +3) — private phase deadline, final leaderboard opening;
- 8 April — paper submission deadline.
Description
Coreference resolution is one of the most difficult and relevant subtasks in the field of natural language processing, important for many applied NLP tasks.
Language expressions referring to the same entity in the world are considered coreferential, for example:
[John] returned home late. [The boy] made excuses for a long time, but in the end he admitted [his] guilt.
Here, the coreference chain is [John] — [the boy] — [his]. Thus, coreference resolution is to make chains of all coreferential references.
Task
We invite participants to solve the problem of coreference resolution for the RuCoCo corpus, consisting of texts from the NewsRU portal.
This coreference resolution competition is the third in the Dialogue Evaluation, but this is the first time a corpus of this size has been used: more than 1 million words just for learning. The RuCoCo corpus was presented at the Dialogue last year.